Column name to column value mapping for a new row (aka data point, observation, sample) to predict.
The purpose in life for objects of type RowData is to be passed to a predict method.
RowData contains the input values for one new row.
In this context, "row" means a new data point (aka row, observation, sample) to make a prediction for.
Column names are mandatory (the column name is the key in the HashMap).
Columns of different types are handled as follows:
-
For numerical columns, the value Object may either be a Double or a String. If a String is passed, then
Double.parseDouble() will be called on the String.
-
For categorical (aka factor, enum) columns, the value Object must be a String with the same names as seen
in the training data.
It is not allowed to use new categorical (aka factor, enum) levels unseen during training (this will result
in a
PredictUnknownCategoricalLevelException when one of the predict methods
is called).
Incorrect use of data types will result in a
PredictUnknownTypeException
when one of the predict methods is called.
For missing columns that are in the model, NA will be used by the predict methods.
Extra columns that are not in the model are ignored by the predict methods.
See the top-of-tree master version of this file
here on github.